





OpenAI launched its Wallet on November 13, 2026. It promised a seamless future for agent-driven commerce. Yet, beneath the polished surface lies an unaddressed liability chasm. The core contradiction is stark. Network giants celebrate frictionless checkouts. They have failed to establish basic dispute rules. Merchants are being asked to absorb risk they cannot audit. The anxiety in payment processing is palpable. Every agent order represents a potential chargeback. It has almost zero evidentiary trail. The industry is marching into undefined territory.
The launch on November 13, 2026, is highly restricted. It covers US Pro subscribers on US-issued cards. Users approve purchases with Face ID or fingerprints. The card number never reaches the shopping agent. The network issues a token scoped to that specific transaction. The system supports Visa, Mastercard, and Amex tokens. Chaya Nayak, OpenAI’s payments lead, focused on merchants. She called the foundation safe and easy. But the token only proves who paid. It does not prove what was agreed. A human checkout leaves a browser session. It also leaves a typed address. An agent checkout leaves a token and a model. There is no dispute reason code for agent purchases. Worldpay’s July 2026 analysis found no formal liability shift. Transactions follow ordinary card-not-present rules. This means merchants face chargebacks with less evidence than usual. The networks have published verification machinery. Visa Intelligent Commerce launched in July 2026. Mastercard Agent Pay introduced agentic tokens. EMVCo published a draft framework in September 2026. None of these tools create a chargeback rule. They only focus on identity and authorization. The practical workaround is delayed captures. Merchants hold captures for fifteen to sixty minutes. They push a notification to the human cardholder. This lets users void the transaction before settlement. But it slows down the checkout process. It also assumes users read their alerts. Forecasts for agentic commerce vary wildly. EMARKETER estimates twenty point five seven billion dollars in 2026. That is only one point five percent of ecommerce. Morgan Stanley projects ten percent of US spending by 2030. Bain counts influenced purchases at up to five hundred billion. The definitions remain highly inconsistent.
The commercial loop is trapped in a Catch-22. For agent commerce to scale, merchants must trust the platform. The networks must codify liability. Yet, the networks have no incentive. They will not shift liability to agents. They profit from transaction volume, regardless of dispute rates. This leaves merchants with a binary choice. They can integrate agent checkout. They must absorb chargeback losses. Or they can bypass agent commerce. They lock themselves out of the next retail wave. The ultimate end-game is a forced consolidation. Payment networks will eventually introduce agent-specific dispute codes. They will only do this once volumes threaten margins. Agent transactions are growing fast. Delayed captures will persist until then. They act as a temporary friction brake. My practical recommendation is simple. Merchants should treat agent orders as high-risk transactions. Do not automate the fulfillment without a human verification step. Keep the delayed capture model. Protect your margins first. The technology is moving too fast for the financial infrastructure.